2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
48 #include "xfs_error.h"
49 #include "xfs_quota.h"
50 #include "xfs_utils.h"
51 #include "xfs_rtalloc.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
55 #include "xfs_vnodeops.h"
56 #include "xfs_trace.h"
64 xfs_mount_t
*mp
= ip
->i_mount
;
65 struct inode
*inode
= VFS_I(ip
);
66 int mask
= iattr
->ia_valid
;
74 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
79 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
80 return XFS_ERROR(EROFS
);
82 if (XFS_FORCED_SHUTDOWN(mp
))
83 return XFS_ERROR(EIO
);
85 code
= -inode_change_ok(inode
, iattr
);
89 olddquot1
= olddquot2
= NULL
;
93 * If disk quotas is on, we make sure that the dquots do exist on disk,
94 * before we start any other transactions. Trying to do this later
95 * is messy. We don't care to take a readlock to look at the ids
96 * in inode here, because we can't hold it across the trans_reserve.
97 * If the IDs do change before we take the ilock, we're covered
98 * because the i_*dquot fields will get updated anyway.
100 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
103 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
105 qflags
|= XFS_QMOPT_UQUOTA
;
107 uid
= ip
->i_d
.di_uid
;
109 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
111 qflags
|= XFS_QMOPT_GQUOTA
;
113 gid
= ip
->i_d
.di_gid
;
117 * We take a reference when we initialize udqp and gdqp,
118 * so it is important that we never blindly double trip on
119 * the same variable. See xfs_create() for an example.
121 ASSERT(udqp
== NULL
);
122 ASSERT(gdqp
== NULL
);
123 code
= xfs_qm_vop_dqalloc(ip
, uid
, gid
, ip
->i_d
.di_projid
,
124 qflags
, &udqp
, &gdqp
);
130 * For the other attributes, we acquire the inode lock and
131 * first do an error checking pass.
134 lock_flags
= XFS_ILOCK_EXCL
;
135 if (flags
& XFS_ATTR_NOLOCK
)
137 if (!(mask
& ATTR_SIZE
)) {
138 if ((mask
!= (ATTR_CTIME
|ATTR_ATIME
|ATTR_MTIME
)) ||
139 (mp
->m_flags
& XFS_MOUNT_WSYNC
)) {
140 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
142 if ((code
= xfs_trans_reserve(tp
, 0,
143 XFS_ICHANGE_LOG_RES(mp
), 0,
150 if (DM_EVENT_ENABLED(ip
, DM_EVENT_TRUNCATE
) &&
151 !(flags
& XFS_ATTR_DMI
)) {
152 int dmflags
= AT_DELAY_FLAG(flags
) | DM_SEM_FLAG_WR
;
153 code
= XFS_SEND_DATA(mp
, DM_EVENT_TRUNCATE
, ip
,
154 iattr
->ia_size
, 0, dmflags
, NULL
);
161 lock_flags
|= XFS_IOLOCK_EXCL
;
164 xfs_ilock(ip
, lock_flags
);
167 * Change file ownership. Must be the owner or privileged.
169 if (mask
& (ATTR_UID
|ATTR_GID
)) {
171 * These IDs could have changed since we last looked at them.
172 * But, we're assured that if the ownership did change
173 * while we didn't have the inode locked, inode's dquot(s)
174 * would have changed also.
176 iuid
= ip
->i_d
.di_uid
;
177 igid
= ip
->i_d
.di_gid
;
178 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
179 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
182 * Do a quota reservation only if uid/gid is actually
185 if (XFS_IS_QUOTA_RUNNING(mp
) &&
186 ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
187 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
))) {
189 code
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
190 capable(CAP_FOWNER
) ?
191 XFS_QMOPT_FORCE_RES
: 0);
192 if (code
) /* out of quota */
198 * Truncate file. Must have write permission and not be a directory.
200 if (mask
& ATTR_SIZE
) {
201 /* Short circuit the truncate case for zero length files */
202 if (iattr
->ia_size
== 0 &&
203 ip
->i_size
== 0 && ip
->i_d
.di_nextents
== 0) {
204 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
205 lock_flags
&= ~XFS_ILOCK_EXCL
;
206 if (mask
& ATTR_CTIME
)
207 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
212 if (S_ISDIR(ip
->i_d
.di_mode
)) {
213 code
= XFS_ERROR(EISDIR
);
215 } else if (!S_ISREG(ip
->i_d
.di_mode
)) {
216 code
= XFS_ERROR(EINVAL
);
221 * Make sure that the dquots are attached to the inode.
223 code
= xfs_qm_dqattach_locked(ip
, 0);
228 * Now we can make the changes. Before we join the inode
229 * to the transaction, if ATTR_SIZE is set then take care of
230 * the part of the truncation that must be done without the
231 * inode lock. This needs to be done before joining the inode
232 * to the transaction, because the inode cannot be unlocked
233 * once it is a part of the transaction.
235 if (iattr
->ia_size
> ip
->i_size
) {
237 * Do the first part of growing a file: zero any data
238 * in the last block that is beyond the old EOF. We
239 * need to do this before the inode is joined to the
240 * transaction to modify the i_size.
242 code
= xfs_zero_eof(ip
, iattr
->ia_size
, ip
->i_size
);
244 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
247 * We are going to log the inode size change in this
248 * transaction so any previous writes that are beyond the on
249 * disk EOF and the new EOF that have not been written out need
250 * to be written here. If we do not write the data out, we
251 * expose ourselves to the null files problem.
253 * Only flush from the on disk size to the smaller of the in
254 * memory file size or the new size as that's the range we
255 * really care about here and prevents waiting for other data
256 * not within the range we care about here.
259 ip
->i_size
!= ip
->i_d
.di_size
&&
260 iattr
->ia_size
> ip
->i_d
.di_size
) {
261 code
= xfs_flush_pages(ip
,
262 ip
->i_d
.di_size
, iattr
->ia_size
,
263 XFS_B_ASYNC
, FI_NONE
);
266 /* wait for all I/O to complete */
270 code
= xfs_itruncate_data(ip
, iattr
->ia_size
);
273 lock_flags
&= ~XFS_ILOCK_EXCL
;
274 ASSERT(lock_flags
== XFS_IOLOCK_EXCL
);
277 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
278 if ((code
= xfs_trans_reserve(tp
, 0,
279 XFS_ITRUNCATE_LOG_RES(mp
), 0,
280 XFS_TRANS_PERM_LOG_RES
,
281 XFS_ITRUNCATE_LOG_COUNT
))) {
282 xfs_trans_cancel(tp
, 0);
284 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
287 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
288 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
290 xfs_trans_ijoin(tp
, ip
, lock_flags
);
291 xfs_trans_ihold(tp
, ip
);
294 * Only change the c/mtime if we are changing the size
295 * or we are explicitly asked to change it. This handles
296 * the semantic difference between truncate() and ftruncate()
297 * as implemented in the VFS.
299 if (iattr
->ia_size
!= ip
->i_size
|| (mask
& ATTR_CTIME
))
300 timeflags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
302 if (iattr
->ia_size
> ip
->i_size
) {
303 ip
->i_d
.di_size
= iattr
->ia_size
;
304 ip
->i_size
= iattr
->ia_size
;
305 if (!(flags
& XFS_ATTR_DMI
))
306 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
307 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
308 } else if (iattr
->ia_size
<= ip
->i_size
||
309 (iattr
->ia_size
== 0 && ip
->i_d
.di_nextents
)) {
311 * signal a sync transaction unless
312 * we're truncating an already unlinked
313 * file on a wsync filesystem
315 code
= xfs_itruncate_finish(&tp
, ip
, iattr
->ia_size
,
317 ((ip
->i_d
.di_nlink
!= 0 ||
318 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
323 * Truncated "down", so we're removing references
324 * to old data here - if we now delay flushing for
325 * a long time, we expose ourselves unduly to the
326 * notorious NULL files problem. So, we mark this
327 * vnode and flush it when the file is closed, and
328 * do not wait the usual (long) time for writeout.
330 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
333 xfs_trans_ijoin(tp
, ip
, lock_flags
);
334 xfs_trans_ihold(tp
, ip
);
338 * Change file ownership. Must be the owner or privileged.
340 if (mask
& (ATTR_UID
|ATTR_GID
)) {
342 * CAP_FSETID overrides the following restrictions:
344 * The set-user-ID and set-group-ID bits of a file will be
345 * cleared upon successful return from chown()
347 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
348 !capable(CAP_FSETID
)) {
349 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
353 * Change the ownerships and register quota modifications
354 * in the transaction.
357 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
358 ASSERT(mask
& ATTR_UID
);
360 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
361 &ip
->i_udquot
, udqp
);
363 ip
->i_d
.di_uid
= uid
;
367 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
368 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
369 ASSERT(mask
& ATTR_GID
);
371 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
372 &ip
->i_gdquot
, gdqp
);
374 ip
->i_d
.di_gid
= gid
;
378 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
379 timeflags
|= XFS_ICHGTIME_CHG
;
383 * Change file access modes.
385 if (mask
& ATTR_MODE
) {
386 umode_t mode
= iattr
->ia_mode
;
388 if (!in_group_p(inode
->i_gid
) && !capable(CAP_FSETID
))
391 ip
->i_d
.di_mode
&= S_IFMT
;
392 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
394 inode
->i_mode
&= S_IFMT
;
395 inode
->i_mode
|= mode
& ~S_IFMT
;
397 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
398 timeflags
|= XFS_ICHGTIME_CHG
;
402 * Change file access or modified times.
404 if (mask
& (ATTR_ATIME
|ATTR_MTIME
)) {
405 if (mask
& ATTR_ATIME
) {
406 inode
->i_atime
= iattr
->ia_atime
;
407 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
408 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
409 ip
->i_update_core
= 1;
411 if (mask
& ATTR_MTIME
) {
412 inode
->i_mtime
= iattr
->ia_mtime
;
413 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
414 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
415 timeflags
&= ~XFS_ICHGTIME_MOD
;
416 timeflags
|= XFS_ICHGTIME_CHG
;
418 if (tp
&& (mask
& (ATTR_MTIME_SET
|ATTR_ATIME_SET
)))
419 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
423 * Change file inode change time only if ATTR_CTIME set
424 * AND we have been called by a DMI function.
427 if ((flags
& XFS_ATTR_DMI
) && (mask
& ATTR_CTIME
)) {
428 inode
->i_ctime
= iattr
->ia_ctime
;
429 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
430 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
431 ip
->i_update_core
= 1;
432 timeflags
&= ~XFS_ICHGTIME_CHG
;
436 * Send out timestamp changes that need to be set to the
437 * current time. Not done when called by a DMI function.
439 if (timeflags
&& !(flags
& XFS_ATTR_DMI
))
440 xfs_ichgtime(ip
, timeflags
);
442 XFS_STATS_INC(xs_ig_attrchg
);
445 * If this is a synchronous mount, make sure that the
446 * transaction goes to disk before returning to the user.
447 * This is slightly sub-optimal in that truncates require
448 * two sync transactions instead of one for wsync filesystems.
449 * One for the truncate and one for the timestamps since we
450 * don't want to change the timestamps unless we're sure the
451 * truncate worked. Truncates are less than 1% of the laddis
452 * mix so this probably isn't worth the trouble to optimize.
456 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
457 xfs_trans_set_sync(tp
);
459 code
= xfs_trans_commit(tp
, commit_flags
);
462 xfs_iunlock(ip
, lock_flags
);
465 * Release any dquot(s) the inode had kept before chown.
467 xfs_qm_dqrele(olddquot1
);
468 xfs_qm_dqrele(olddquot2
);
476 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
477 * update. We could avoid this with linked transactions
478 * and passing down the transaction pointer all the way
479 * to attr_set. No previous user of the generic
480 * Posix ACL code seems to care about this issue either.
482 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
483 code
= -xfs_acl_chmod(inode
);
485 return XFS_ERROR(code
);
488 if (DM_EVENT_ENABLED(ip
, DM_EVENT_ATTRIBUTE
) &&
489 !(flags
& XFS_ATTR_DMI
)) {
490 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_ATTRIBUTE
, ip
, DM_RIGHT_NULL
,
491 NULL
, DM_RIGHT_NULL
, NULL
, NULL
,
492 0, 0, AT_DELAY_FLAG(flags
));
497 commit_flags
|= XFS_TRANS_ABORT
;
503 xfs_trans_cancel(tp
, commit_flags
);
505 if (lock_flags
!= 0) {
506 xfs_iunlock(ip
, lock_flags
);
512 * The maximum pathlen is 1024 bytes. Since the minimum file system
513 * blocksize is 512 bytes, we can get a max of 2 extents back from
516 #define SYMLINK_MAPS 2
523 xfs_mount_t
*mp
= ip
->i_mount
;
524 int pathlen
= ip
->i_d
.di_size
;
525 int nmaps
= SYMLINK_MAPS
;
526 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
533 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
), 0, NULL
, 0,
534 mval
, &nmaps
, NULL
, NULL
);
538 for (n
= 0; n
< nmaps
; n
++) {
539 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
540 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
542 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
, BTOBB(byte_cnt
),
543 XBF_LOCK
| XBF_MAPPED
| XBF_DONT_BLOCK
);
544 error
= XFS_BUF_GETERROR(bp
);
546 xfs_ioerror_alert("xfs_readlink",
547 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
551 if (pathlen
< byte_cnt
)
555 memcpy(link
, XFS_BUF_PTR(bp
), byte_cnt
);
559 link
[ip
->i_d
.di_size
] = '\0';
571 xfs_mount_t
*mp
= ip
->i_mount
;
575 xfs_itrace_entry(ip
);
577 if (XFS_FORCED_SHUTDOWN(mp
))
578 return XFS_ERROR(EIO
);
580 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
582 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
);
583 ASSERT(ip
->i_d
.di_size
<= MAXPATHLEN
);
585 pathlen
= ip
->i_d
.di_size
;
589 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
590 memcpy(link
, ip
->i_df
.if_u1
.if_data
, pathlen
);
591 link
[pathlen
] = '\0';
593 error
= xfs_readlink_bmap(ip
, link
);
597 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
604 * This is called to sync the inode and its data out to disk. We need to hold
605 * the I/O lock while flushing the data, and the inode lock while flushing the
606 * inode. The inode lock CANNOT be held while flushing the data, so acquire
607 * after we're done with that.
615 int log_flushed
= 0, changed
= 1;
617 xfs_itrace_entry(ip
);
619 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
620 return XFS_ERROR(EIO
);
623 * We always need to make sure that the required inode state is safe on
624 * disk. The inode might be clean but we still might need to force the
625 * log because of committed transactions that haven't hit the disk yet.
626 * Likewise, there could be unflushed non-transactional changes to the
627 * inode core that have to go to disk and this requires us to issue
628 * a synchronous transaction to capture these changes correctly.
630 * This code relies on the assumption that if the update_* fields
631 * of the inode are clear and the inode is unpinned then it is clean
632 * and no action is required.
634 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
636 if (!ip
->i_update_core
) {
638 * Timestamps/size haven't changed since last inode flush or
639 * inode transaction commit. That means either nothing got
640 * written or a transaction committed which caught the updates.
641 * If the latter happened and the transaction hasn't hit the
642 * disk yet, the inode will be still be pinned. If it is,
646 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
648 if (xfs_ipincount(ip
)) {
649 error
= _xfs_log_force(ip
->i_mount
, (xfs_lsn_t
)0,
650 XFS_LOG_FORCE
| XFS_LOG_SYNC
,
654 * If the inode is not pinned and nothing has changed
655 * we don't need to flush the cache.
661 * Kick off a transaction to log the inode core to get the
662 * updates. The sync transaction will also force the log.
664 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
665 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
666 error
= xfs_trans_reserve(tp
, 0,
667 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
), 0, 0, 0);
669 xfs_trans_cancel(tp
, 0);
672 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
675 * Note - it's possible that we might have pushed ourselves out
676 * of the way during trans_reserve which would flush the inode.
677 * But there's no guarantee that the inode buffer has actually
678 * gone out yet (it's delwri). Plus the buffer could be pinned
679 * anyway if it's part of an inode in another recent
680 * transaction. So we play it safe and fire off the
681 * transaction anyway.
683 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
684 xfs_trans_ihold(tp
, ip
);
685 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
686 xfs_trans_set_sync(tp
);
687 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
689 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
692 if ((ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) && changed
) {
694 * If the log write didn't issue an ordered tag we need
695 * to flush the disk cache for the data device now.
698 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
701 * If this inode is on the RT dev we need to flush that
704 if (XFS_IS_REALTIME_INODE(ip
))
705 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
712 * Flags for xfs_free_eofblocks
714 #define XFS_FREE_EOF_TRYLOCK (1<<0)
717 * This is called by xfs_inactive to free any blocks beyond eof
718 * when the link count isn't zero and by xfs_dm_punch_hole() when
719 * punching a hole to EOF.
729 xfs_fileoff_t end_fsb
;
730 xfs_fileoff_t last_fsb
;
731 xfs_filblks_t map_len
;
733 xfs_bmbt_irec_t imap
;
736 * Figure out if there are any blocks beyond the end
737 * of the file. If not, then there is nothing to do.
739 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_size
));
740 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
741 map_len
= last_fsb
- end_fsb
;
746 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
747 error
= xfs_bmapi(NULL
, ip
, end_fsb
, map_len
, 0,
748 NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
749 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
751 if (!error
&& (nimaps
!= 0) &&
752 (imap
.br_startblock
!= HOLESTARTBLOCK
||
753 ip
->i_delayed_blks
)) {
755 * Attach the dquots to the inode up front.
757 error
= xfs_qm_dqattach(ip
, 0);
762 * There are blocks after the end of file.
763 * Free them up now by truncating the file to
766 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
769 * Do the xfs_itruncate_start() call before
770 * reserving any log space because
771 * itruncate_start will call into the buffer
773 * do that within a transaction.
775 if (flags
& XFS_FREE_EOF_TRYLOCK
) {
776 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
777 xfs_trans_cancel(tp
, 0);
781 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
783 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
786 xfs_trans_cancel(tp
, 0);
787 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
791 error
= xfs_trans_reserve(tp
, 0,
792 XFS_ITRUNCATE_LOG_RES(mp
),
793 0, XFS_TRANS_PERM_LOG_RES
,
794 XFS_ITRUNCATE_LOG_COUNT
);
796 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
797 xfs_trans_cancel(tp
, 0);
798 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
802 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
803 xfs_trans_ijoin(tp
, ip
,
806 xfs_trans_ihold(tp
, ip
);
808 error
= xfs_itruncate_finish(&tp
, ip
,
813 * If we get an error at this point we
814 * simply don't bother truncating the file.
818 (XFS_TRANS_RELEASE_LOG_RES
|
821 error
= xfs_trans_commit(tp
,
822 XFS_TRANS_RELEASE_LOG_RES
);
824 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
);
830 * Free a symlink that has blocks associated with it.
833 xfs_inactive_symlink_rmt(
841 xfs_fsblock_t first_block
;
842 xfs_bmap_free_t free_list
;
845 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
853 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
855 * We're freeing a symlink that has some
856 * blocks allocated to it. Free the
857 * blocks here. We know that we've got
858 * either 1 or 2 extents and that we can
859 * free them all in one bunmapi call.
861 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
862 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
863 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
864 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
865 xfs_trans_cancel(tp
, 0);
870 * Lock the inode, fix the size, and join it to the transaction.
871 * Hold it so in the normal path, we still have it locked for
872 * the second transaction. In the error paths we need it
873 * held so the cancel won't rele it, see below.
875 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
876 size
= (int)ip
->i_d
.di_size
;
878 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
879 xfs_trans_ihold(tp
, ip
);
880 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
882 * Find the block(s) so we can inval and unmap them.
885 xfs_bmap_init(&free_list
, &first_block
);
886 nmaps
= ARRAY_SIZE(mval
);
887 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
888 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
892 * Invalidate the block(s).
894 for (i
= 0; i
< nmaps
; i
++) {
895 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
896 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
897 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
898 xfs_trans_binval(tp
, bp
);
901 * Unmap the dead block(s) to the free_list.
903 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
904 &first_block
, &free_list
, NULL
, &done
)))
908 * Commit the first transaction. This logs the EFI and the inode.
910 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
913 * The transaction must have been committed, since there were
914 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
915 * The new tp has the extent freeing and EFDs.
919 * The first xact was committed, so add the inode to the new one.
920 * Mark it dirty so it will be logged and moved forward in the log as
921 * part of every commit.
923 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
924 xfs_trans_ihold(tp
, ip
);
925 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
927 * Get a new, empty transaction to return to our caller.
929 ntp
= xfs_trans_dup(tp
);
931 * Commit the transaction containing extent freeing and EFDs.
932 * If we get an error on the commit here or on the reserve below,
933 * we need to unlock the inode since the new transaction doesn't
934 * have the inode attached.
936 error
= xfs_trans_commit(tp
, 0);
939 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
943 * transaction commit worked ok so we can drop the extra ticket
944 * reference that we gained in xfs_trans_dup()
946 xfs_log_ticket_put(tp
->t_ticket
);
949 * Remove the memory for extent descriptions (just bookkeeping).
951 if (ip
->i_df
.if_bytes
)
952 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
953 ASSERT(ip
->i_df
.if_bytes
== 0);
955 * Put an itruncate log reservation in the new transaction
958 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
959 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
960 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
964 * Return with the inode locked but not joined to the transaction.
970 xfs_bmap_cancel(&free_list
);
973 * Have to come here with the inode locked and either
974 * (held and in the transaction) or (not in the transaction).
975 * If the inode isn't held then cancel would iput it, but
976 * that's wrong since this is inactive and the vnode ref
977 * count is 0 already.
978 * Cancel won't do anything to the inode if held, but it still
979 * needs to be locked until the cancel is done, if it was
980 * joined to the transaction.
982 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
983 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
990 xfs_inactive_symlink_local(
996 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
998 * We're freeing a symlink which fit into
999 * the inode. Just free the memory used
1000 * to hold the old symlink.
1002 error
= xfs_trans_reserve(*tpp
, 0,
1003 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
1004 0, XFS_TRANS_PERM_LOG_RES
,
1005 XFS_ITRUNCATE_LOG_COUNT
);
1008 xfs_trans_cancel(*tpp
, 0);
1012 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1015 * Zero length symlinks _can_ exist.
1017 if (ip
->i_df
.if_bytes
> 0) {
1018 xfs_idata_realloc(ip
,
1019 -(ip
->i_df
.if_bytes
),
1021 ASSERT(ip
->i_df
.if_bytes
== 0);
1035 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1038 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1039 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1040 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1044 error
= xfs_attr_inactive(ip
);
1048 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1049 error
= xfs_trans_reserve(tp
, 0,
1050 XFS_IFREE_LOG_RES(mp
),
1051 0, XFS_TRANS_PERM_LOG_RES
,
1052 XFS_INACTIVE_LOG_COUNT
);
1056 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1057 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1058 xfs_trans_ihold(tp
, ip
);
1059 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1061 ASSERT(ip
->i_d
.di_anextents
== 0);
1067 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1068 xfs_trans_cancel(tp
, 0);
1071 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1079 xfs_mount_t
*mp
= ip
->i_mount
;
1082 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1085 /* If this is a read-only mount, don't do this (would generate I/O) */
1086 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1089 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1093 * If we are using filestreams, and we have an unlinked
1094 * file that we are processing the last close on, then nothing
1095 * will be able to reopen and write to this file. Purge this
1096 * inode from the filestreams cache so that it doesn't delay
1097 * teardown of the inode.
1099 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1100 xfs_filestream_deassociate(ip
);
1103 * If we previously truncated this file and removed old data
1104 * in the process, we want to initiate "early" writeout on
1105 * the last close. This is an attempt to combat the notorious
1106 * NULL files problem which is particularly noticable from a
1107 * truncate down, buffered (re-)write (delalloc), followed by
1108 * a crash. What we are effectively doing here is
1109 * significantly reducing the time window where we'd otherwise
1110 * be exposed to that problem.
1112 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1113 if (truncated
&& VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0)
1114 xfs_flush_pages(ip
, 0, -1, XFS_B_ASYNC
, FI_NONE
);
1117 if (ip
->i_d
.di_nlink
!= 0) {
1118 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1119 ((ip
->i_size
> 0) || (VN_CACHED(VFS_I(ip
)) > 0 ||
1120 ip
->i_delayed_blks
> 0)) &&
1121 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
1122 (!(ip
->i_d
.di_flags
&
1123 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
1126 * If we can't get the iolock just skip truncating
1127 * the blocks past EOF because we could deadlock
1128 * with the mmap_sem otherwise. We'll get another
1129 * chance to drop them once the last reference to
1130 * the inode is dropped, so we'll never leak blocks
1133 error
= xfs_free_eofblocks(mp
, ip
,
1134 XFS_FREE_EOF_TRYLOCK
);
1146 * This is called when the vnode reference count for the vnode
1147 * goes to zero. If the file has been unlinked, then it must
1148 * now be truncated. Also, we clear all of the read-ahead state
1149 * kept for the inode here since the file is now closed.
1155 xfs_bmap_free_t free_list
;
1156 xfs_fsblock_t first_block
;
1163 xfs_itrace_entry(ip
);
1166 * If the inode is already free, then there can be nothing
1169 if (ip
->i_d
.di_mode
== 0 || is_bad_inode(VFS_I(ip
))) {
1170 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1171 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1172 return VN_INACTIVE_CACHE
;
1176 * Only do a truncate if it's a regular file with
1177 * some actual space in it. It's OK to look at the
1178 * inode's fields without the lock because we're the
1179 * only one with a reference to the inode.
1181 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1182 ((ip
->i_d
.di_size
!= 0) || (ip
->i_size
!= 0) ||
1183 (ip
->i_d
.di_nextents
> 0) || (ip
->i_delayed_blks
> 0)) &&
1184 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1188 if (ip
->i_d
.di_nlink
== 0 && DM_EVENT_ENABLED(ip
, DM_EVENT_DESTROY
))
1189 XFS_SEND_DESTROY(mp
, ip
, DM_RIGHT_NULL
);
1193 /* If this is a read-only mount, don't do this (would generate I/O) */
1194 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1197 if (ip
->i_d
.di_nlink
!= 0) {
1198 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1199 ((ip
->i_size
> 0) || (VN_CACHED(VFS_I(ip
)) > 0 ||
1200 ip
->i_delayed_blks
> 0)) &&
1201 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1202 (!(ip
->i_d
.di_flags
&
1203 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1204 (ip
->i_delayed_blks
!= 0)))) {
1205 error
= xfs_free_eofblocks(mp
, ip
, 0);
1207 return VN_INACTIVE_CACHE
;
1212 ASSERT(ip
->i_d
.di_nlink
== 0);
1214 error
= xfs_qm_dqattach(ip
, 0);
1216 return VN_INACTIVE_CACHE
;
1218 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1221 * Do the xfs_itruncate_start() call before
1222 * reserving any log space because itruncate_start
1223 * will call into the buffer cache and we can't
1224 * do that within a transaction.
1226 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1228 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1230 xfs_trans_cancel(tp
, 0);
1231 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1232 return VN_INACTIVE_CACHE
;
1235 error
= xfs_trans_reserve(tp
, 0,
1236 XFS_ITRUNCATE_LOG_RES(mp
),
1237 0, XFS_TRANS_PERM_LOG_RES
,
1238 XFS_ITRUNCATE_LOG_COUNT
);
1240 /* Don't call itruncate_cleanup */
1241 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1242 xfs_trans_cancel(tp
, 0);
1243 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1244 return VN_INACTIVE_CACHE
;
1247 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1248 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1249 xfs_trans_ihold(tp
, ip
);
1252 * normally, we have to run xfs_itruncate_finish sync.
1253 * But if filesystem is wsync and we're in the inactive
1254 * path, then we know that nlink == 0, and that the
1255 * xaction that made nlink == 0 is permanently committed
1256 * since xfs_remove runs as a synchronous transaction.
1258 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1259 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1262 xfs_trans_cancel(tp
,
1263 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1264 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1265 return VN_INACTIVE_CACHE
;
1267 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1270 * If we get an error while cleaning up a
1271 * symlink we bail out.
1273 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1274 xfs_inactive_symlink_rmt(ip
, &tp
) :
1275 xfs_inactive_symlink_local(ip
, &tp
);
1279 return VN_INACTIVE_CACHE
;
1282 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1283 xfs_trans_ihold(tp
, ip
);
1285 error
= xfs_trans_reserve(tp
, 0,
1286 XFS_IFREE_LOG_RES(mp
),
1287 0, XFS_TRANS_PERM_LOG_RES
,
1288 XFS_INACTIVE_LOG_COUNT
);
1290 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1291 xfs_trans_cancel(tp
, 0);
1292 return VN_INACTIVE_CACHE
;
1295 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1296 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1297 xfs_trans_ihold(tp
, ip
);
1301 * If there are attributes associated with the file
1302 * then blow them away now. The code calls a routine
1303 * that recursively deconstructs the attribute fork.
1304 * We need to just commit the current transaction
1305 * because we can't use it for xfs_attr_inactive().
1307 if (ip
->i_d
.di_anextents
> 0) {
1308 error
= xfs_inactive_attrs(ip
, &tp
);
1310 * If we got an error, the transaction is already
1311 * cancelled, and the inode is unlocked. Just get out.
1314 return VN_INACTIVE_CACHE
;
1315 } else if (ip
->i_afp
) {
1316 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1322 xfs_bmap_init(&free_list
, &first_block
);
1323 error
= xfs_ifree(tp
, ip
, &free_list
);
1326 * If we fail to free the inode, shut down. The cancel
1327 * might do that, we need to make sure. Otherwise the
1328 * inode might be lost for a long time or forever.
1330 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1332 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1333 error
, mp
->m_fsname
);
1334 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1336 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1339 * Credit the quota account(s). The inode is gone.
1341 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1344 * Just ignore errors at this point. There is nothing we can
1345 * do except to try to keep going. Make sure it's not a silent
1348 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1350 xfs_fs_cmn_err(CE_NOTE
, mp
, "xfs_inactive: "
1351 "xfs_bmap_finish() returned error %d", error
);
1352 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1354 xfs_fs_cmn_err(CE_NOTE
, mp
, "xfs_inactive: "
1355 "xfs_trans_commit() returned error %d", error
);
1359 * Release the dquots held by inode, if any.
1361 xfs_qm_dqdetach(ip
);
1362 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1365 return VN_INACTIVE_CACHE
;
1369 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1370 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1371 * ci_name->name will point to a the actual name (caller must free) or
1372 * will be set to NULL if an exact match is found.
1377 struct xfs_name
*name
,
1379 struct xfs_name
*ci_name
)
1385 xfs_itrace_entry(dp
);
1387 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1388 return XFS_ERROR(EIO
);
1390 lock_mode
= xfs_ilock_map_shared(dp
);
1391 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
1392 xfs_iunlock_map_shared(dp
, lock_mode
);
1397 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
, 0);
1405 kmem_free(ci_name
->name
);
1414 struct xfs_name
*name
,
1420 int is_dir
= S_ISDIR(mode
);
1421 struct xfs_mount
*mp
= dp
->i_mount
;
1422 struct xfs_inode
*ip
= NULL
;
1423 struct xfs_trans
*tp
= NULL
;
1425 xfs_bmap_free_t free_list
;
1426 xfs_fsblock_t first_block
;
1427 boolean_t unlock_dp_on_error
= B_FALSE
;
1431 struct xfs_dquot
*udqp
= NULL
;
1432 struct xfs_dquot
*gdqp
= NULL
;
1437 xfs_itrace_entry(dp
);
1439 if (XFS_FORCED_SHUTDOWN(mp
))
1440 return XFS_ERROR(EIO
);
1442 if (DM_EVENT_ENABLED(dp
, DM_EVENT_CREATE
)) {
1443 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
1444 dp
, DM_RIGHT_NULL
, NULL
,
1445 DM_RIGHT_NULL
, name
->name
, NULL
,
1452 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1453 prid
= dp
->i_d
.di_projid
;
1458 * Make sure that we have allocated dquot(s) on disk.
1460 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1461 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1467 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1468 log_res
= XFS_MKDIR_LOG_RES(mp
);
1469 log_count
= XFS_MKDIR_LOG_COUNT
;
1470 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1472 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1473 log_res
= XFS_CREATE_LOG_RES(mp
);
1474 log_count
= XFS_CREATE_LOG_COUNT
;
1475 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1478 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1481 * Initially assume that the file does not exist and
1482 * reserve the resources for that case. If that is not
1483 * the case we'll drop the one we have and get a more
1484 * appropriate transaction later.
1486 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
1487 XFS_TRANS_PERM_LOG_RES
, log_count
);
1488 if (error
== ENOSPC
) {
1489 /* flush outstanding delalloc blocks and retry */
1490 xfs_flush_inodes(dp
);
1491 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
1492 XFS_TRANS_PERM_LOG_RES
, log_count
);
1494 if (error
== ENOSPC
) {
1495 /* No space at all so try a "no-allocation" reservation */
1497 error
= xfs_trans_reserve(tp
, 0, log_res
, 0,
1498 XFS_TRANS_PERM_LOG_RES
, log_count
);
1502 goto out_trans_cancel
;
1505 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1506 unlock_dp_on_error
= B_TRUE
;
1509 * Check for directory link count overflow.
1511 if (is_dir
&& dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
1512 error
= XFS_ERROR(EMLINK
);
1513 goto out_trans_cancel
;
1516 xfs_bmap_init(&free_list
, &first_block
);
1519 * Reserve disk quota and the inode.
1521 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
1523 goto out_trans_cancel
;
1525 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
1527 goto out_trans_cancel
;
1530 * A newly created regular or special file just has one directory
1531 * entry pointing to them, but a directory also the "." entry
1532 * pointing to itself.
1534 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
, credp
,
1535 prid
, resblks
> 0, &ip
, &committed
);
1537 if (error
== ENOSPC
)
1538 goto out_trans_cancel
;
1539 goto out_trans_abort
;
1543 * At this point, we've gotten a newly allocated inode.
1544 * It is locked (and joined to the transaction).
1546 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1549 * Now we join the directory inode to the transaction. We do not do it
1550 * earlier because xfs_dir_ialloc might commit the previous transaction
1551 * (and release all the locks). An error from here on will result in
1552 * the transaction cancel unlocking dp so don't do it explicitly in the
1556 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1557 unlock_dp_on_error
= B_FALSE
;
1559 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1560 &first_block
, &free_list
, resblks
?
1561 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1563 ASSERT(error
!= ENOSPC
);
1564 goto out_trans_abort
;
1566 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1567 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1570 error
= xfs_dir_init(tp
, ip
, dp
);
1572 goto out_bmap_cancel
;
1574 error
= xfs_bumplink(tp
, dp
);
1576 goto out_bmap_cancel
;
1580 * If this is a synchronous mount, make sure that the
1581 * create transaction goes to disk before returning to
1584 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1585 xfs_trans_set_sync(tp
);
1588 * Attach the dquot(s) to the inodes and modify them incore.
1589 * These ids of the inode couldn't have changed since the new
1590 * inode has been locked ever since it was created.
1592 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
1595 * xfs_trans_commit normally decrements the vnode ref count
1596 * when it unlocks the inode. Since we want to return the
1597 * vnode to the caller, we bump the vnode ref count now.
1601 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1603 goto out_abort_rele
;
1605 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1611 xfs_qm_dqrele(udqp
);
1612 xfs_qm_dqrele(gdqp
);
1616 /* Fallthrough to std_return with error = 0 */
1618 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTCREATE
)) {
1619 XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
, dp
, DM_RIGHT_NULL
,
1620 ip
, DM_RIGHT_NULL
, name
->name
, NULL
, mode
,
1627 xfs_bmap_cancel(&free_list
);
1629 cancel_flags
|= XFS_TRANS_ABORT
;
1631 xfs_trans_cancel(tp
, cancel_flags
);
1633 xfs_qm_dqrele(udqp
);
1634 xfs_qm_dqrele(gdqp
);
1636 if (unlock_dp_on_error
)
1637 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1643 * Wait until after the current transaction is aborted to
1644 * release the inode. This prevents recursive transactions
1645 * and deadlocks from xfs_inactive.
1647 xfs_bmap_cancel(&free_list
);
1648 cancel_flags
|= XFS_TRANS_ABORT
;
1649 xfs_trans_cancel(tp
, cancel_flags
);
1651 unlock_dp_on_error
= B_FALSE
;
1657 int xfs_small_retries
;
1658 int xfs_middle_retries
;
1659 int xfs_lots_retries
;
1660 int xfs_lock_delays
;
1664 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1668 xfs_lock_inumorder(int lock_mode
, int subclass
)
1670 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1671 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
1672 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
1673 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
1679 * The following routine will lock n inodes in exclusive mode.
1680 * We assume the caller calls us with the inodes in i_ino order.
1682 * We need to detect deadlock where an inode that we lock
1683 * is in the AIL and we start waiting for another inode that is locked
1684 * by a thread in a long running transaction (such as truncate). This can
1685 * result in deadlock since the long running trans might need to wait
1686 * for the inode we just locked in order to push the tail and free space
1695 int attempts
= 0, i
, j
, try_lock
;
1698 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
1704 for (; i
< inodes
; i
++) {
1707 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
1711 * If try_lock is not set yet, make sure all locked inodes
1712 * are not in the AIL.
1713 * If any are, set try_lock to be used later.
1717 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
1718 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
1719 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1726 * If any of the previous locks we have locked is in the AIL,
1727 * we must TRY to get the second and subsequent locks. If
1728 * we can't get any, we must release all we have
1733 /* try_lock must be 0 if i is 0. */
1735 * try_lock means we have an inode locked
1736 * that is in the AIL.
1739 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
1743 * Unlock all previous guys and try again.
1744 * xfs_iunlock will try to push the tail
1745 * if the inode is in the AIL.
1748 for(j
= i
- 1; j
>= 0; j
--) {
1751 * Check to see if we've already
1752 * unlocked this one.
1753 * Not the first one going back,
1754 * and the inode ptr is the same.
1756 if ((j
!= (i
- 1)) && ips
[j
] ==
1760 xfs_iunlock(ips
[j
], lock_mode
);
1763 if ((attempts
% 5) == 0) {
1764 delay(1); /* Don't just spin the CPU */
1774 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
1780 if (attempts
< 5) xfs_small_retries
++;
1781 else if (attempts
< 100) xfs_middle_retries
++;
1782 else xfs_lots_retries
++;
1790 * xfs_lock_two_inodes() can only be used to lock one type of lock
1791 * at a time - the iolock or the ilock, but not both at once. If
1792 * we lock both at once, lockdep will report false positives saying
1793 * we have violated locking orders.
1796 xfs_lock_two_inodes(
1805 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1806 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
1807 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
1809 if (ip0
->i_ino
> ip1
->i_ino
) {
1816 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
1819 * If the first lock we have locked is in the AIL, we must TRY to get
1820 * the second lock. If we can't get it, we must release the first one
1823 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
1824 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1825 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
1826 xfs_iunlock(ip0
, lock_mode
);
1827 if ((++attempts
% 5) == 0)
1828 delay(1); /* Don't just spin the CPU */
1832 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
1839 struct xfs_name
*name
,
1842 xfs_mount_t
*mp
= dp
->i_mount
;
1843 xfs_trans_t
*tp
= NULL
;
1844 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
1846 xfs_bmap_free_t free_list
;
1847 xfs_fsblock_t first_block
;
1854 xfs_itrace_entry(dp
);
1855 xfs_itrace_entry(ip
);
1857 if (XFS_FORCED_SHUTDOWN(mp
))
1858 return XFS_ERROR(EIO
);
1860 if (DM_EVENT_ENABLED(dp
, DM_EVENT_REMOVE
)) {
1861 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
, dp
, DM_RIGHT_NULL
,
1862 NULL
, DM_RIGHT_NULL
, name
->name
, NULL
,
1863 ip
->i_d
.di_mode
, 0, 0);
1868 error
= xfs_qm_dqattach(dp
, 0);
1872 error
= xfs_qm_dqattach(ip
, 0);
1877 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
1878 log_count
= XFS_DEFAULT_LOG_COUNT
;
1880 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
1881 log_count
= XFS_REMOVE_LOG_COUNT
;
1883 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1886 * We try to get the real space reservation first,
1887 * allowing for directory btree deletion(s) implying
1888 * possible bmap insert(s). If we can't get the space
1889 * reservation then we use 0 instead, and avoid the bmap
1890 * btree insert(s) in the directory code by, if the bmap
1891 * insert tries to happen, instead trimming the LAST
1892 * block from the directory.
1894 resblks
= XFS_REMOVE_SPACE_RES(mp
);
1895 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
1896 XFS_TRANS_PERM_LOG_RES
, log_count
);
1897 if (error
== ENOSPC
) {
1899 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
1900 XFS_TRANS_PERM_LOG_RES
, log_count
);
1903 ASSERT(error
!= ENOSPC
);
1905 goto out_trans_cancel
;
1908 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
1911 * At this point, we've gotten both the directory and the entry
1915 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1918 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1921 * If we're removing a directory perform some additional validation.
1924 ASSERT(ip
->i_d
.di_nlink
>= 2);
1925 if (ip
->i_d
.di_nlink
!= 2) {
1926 error
= XFS_ERROR(ENOTEMPTY
);
1927 goto out_trans_cancel
;
1929 if (!xfs_dir_isempty(ip
)) {
1930 error
= XFS_ERROR(ENOTEMPTY
);
1931 goto out_trans_cancel
;
1935 xfs_bmap_init(&free_list
, &first_block
);
1936 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
1937 &first_block
, &free_list
, resblks
);
1939 ASSERT(error
!= ENOENT
);
1940 goto out_bmap_cancel
;
1942 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1946 * Drop the link from ip's "..".
1948 error
= xfs_droplink(tp
, dp
);
1950 goto out_bmap_cancel
;
1953 * Drop the "." link from ip to self.
1955 error
= xfs_droplink(tp
, ip
);
1957 goto out_bmap_cancel
;
1960 * When removing a non-directory we need to log the parent
1961 * inode here. For a directory this is done implicitly
1962 * by the xfs_droplink call for the ".." entry.
1964 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1968 * Drop the link from dp to ip.
1970 error
= xfs_droplink(tp
, ip
);
1972 goto out_bmap_cancel
;
1975 * Determine if this is the last link while
1976 * we are in the transaction.
1978 link_zero
= (ip
->i_d
.di_nlink
== 0);
1981 * If this is a synchronous mount, make sure that the
1982 * remove transaction goes to disk before returning to
1985 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1986 xfs_trans_set_sync(tp
);
1988 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1990 goto out_bmap_cancel
;
1992 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1997 * If we are using filestreams, kill the stream association.
1998 * If the file is still open it may get a new one but that
1999 * will get killed on last close in xfs_close() so we don't
2000 * have to worry about that.
2002 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
2003 xfs_filestream_deassociate(ip
);
2006 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTREMOVE
)) {
2007 XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
, dp
, DM_RIGHT_NULL
,
2008 NULL
, DM_RIGHT_NULL
, name
->name
, NULL
,
2009 ip
->i_d
.di_mode
, error
, 0);
2015 xfs_bmap_cancel(&free_list
);
2016 cancel_flags
|= XFS_TRANS_ABORT
;
2018 xfs_trans_cancel(tp
, cancel_flags
);
2026 struct xfs_name
*target_name
)
2028 xfs_mount_t
*mp
= tdp
->i_mount
;
2031 xfs_bmap_free_t free_list
;
2032 xfs_fsblock_t first_block
;
2037 xfs_itrace_entry(tdp
);
2038 xfs_itrace_entry(sip
);
2040 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
2042 if (XFS_FORCED_SHUTDOWN(mp
))
2043 return XFS_ERROR(EIO
);
2045 if (DM_EVENT_ENABLED(tdp
, DM_EVENT_LINK
)) {
2046 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_LINK
,
2049 target_name
->name
, NULL
, 0, 0, 0);
2054 /* Return through std_return after this point. */
2056 error
= xfs_qm_dqattach(sip
, 0);
2060 error
= xfs_qm_dqattach(tdp
, 0);
2064 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
2065 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2066 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
2067 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
2068 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2069 if (error
== ENOSPC
) {
2071 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
2072 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2079 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
2082 * Increment vnode ref counts since xfs_trans_commit &
2083 * xfs_trans_cancel will both unlock the inodes and
2084 * decrement the associated ref counts.
2088 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
2089 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
2092 * If the source has too many links, we can't make any more to it.
2094 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2095 error
= XFS_ERROR(EMLINK
);
2100 * If we are using project inheritance, we only allow hard link
2101 * creation in our tree when the project IDs are the same; else
2102 * the tree quota mechanism could be circumvented.
2104 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2105 (tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
))) {
2106 error
= XFS_ERROR(EXDEV
);
2110 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
2114 xfs_bmap_init(&free_list
, &first_block
);
2116 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
2117 &first_block
, &free_list
, resblks
);
2120 xfs_ichgtime(tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2121 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
2123 error
= xfs_bumplink(tp
, sip
);
2128 * If this is a synchronous mount, make sure that the
2129 * link transaction goes to disk before returning to
2132 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2133 xfs_trans_set_sync(tp
);
2136 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
2138 xfs_bmap_cancel(&free_list
);
2142 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2146 /* Fall through to std_return with error = 0. */
2148 if (DM_EVENT_ENABLED(sip
, DM_EVENT_POSTLINK
)) {
2149 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTLINK
,
2152 target_name
->name
, NULL
, 0, error
, 0);
2157 cancel_flags
|= XFS_TRANS_ABORT
;
2161 xfs_trans_cancel(tp
, cancel_flags
);
2168 struct xfs_name
*link_name
,
2169 const char *target_path
,
2174 xfs_mount_t
*mp
= dp
->i_mount
;
2179 xfs_bmap_free_t free_list
;
2180 xfs_fsblock_t first_block
;
2181 boolean_t unlock_dp_on_error
= B_FALSE
;
2184 xfs_fileoff_t first_fsb
;
2185 xfs_filblks_t fs_blocks
;
2187 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
2189 const char *cur_chunk
;
2194 struct xfs_dquot
*udqp
, *gdqp
;
2202 xfs_itrace_entry(dp
);
2204 if (XFS_FORCED_SHUTDOWN(mp
))
2205 return XFS_ERROR(EIO
);
2208 * Check component lengths of the target path name.
2210 pathlen
= strlen(target_path
);
2211 if (pathlen
>= MAXPATHLEN
) /* total string too long */
2212 return XFS_ERROR(ENAMETOOLONG
);
2214 if (DM_EVENT_ENABLED(dp
, DM_EVENT_SYMLINK
)) {
2215 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_SYMLINK
, dp
,
2216 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
2217 link_name
->name
, target_path
, 0, 0, 0);
2222 /* Return through std_return after this point. */
2225 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
2226 prid
= dp
->i_d
.di_projid
;
2228 prid
= (xfs_prid_t
)dfltprid
;
2231 * Make sure that we have allocated dquot(s) on disk.
2233 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
2234 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2238 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
2239 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2241 * The symlink will fit into the inode data fork?
2242 * There can't be any attributes so we get the whole variable part.
2244 if (pathlen
<= XFS_LITINO(mp
))
2247 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
2248 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_name
->len
, fs_blocks
);
2249 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
2250 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
2251 if (error
== ENOSPC
&& fs_blocks
== 0) {
2253 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
2254 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
2261 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2262 unlock_dp_on_error
= B_TRUE
;
2265 * Check whether the directory allows new symlinks or not.
2267 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
2268 error
= XFS_ERROR(EPERM
);
2273 * Reserve disk quota : blocks and inode.
2275 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
2280 * Check for ability to enter directory entry, if no space reserved.
2282 error
= xfs_dir_canenter(tp
, dp
, link_name
, resblks
);
2286 * Initialize the bmap freelist prior to calling either
2287 * bmapi or the directory create code.
2289 xfs_bmap_init(&free_list
, &first_block
);
2292 * Allocate an inode for the symlink.
2294 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (mode
& ~S_IFMT
),
2295 1, 0, credp
, prid
, resblks
> 0, &ip
, NULL
);
2297 if (error
== ENOSPC
)
2303 * An error after we've joined dp to the transaction will result in the
2304 * transaction cancel unlocking dp so don't do it explicitly in the
2308 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2309 unlock_dp_on_error
= B_FALSE
;
2312 * Also attach the dquot(s) to it, if applicable.
2314 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
2317 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
2319 * If the symlink will fit into the inode, write it inline.
2321 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
2322 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
2323 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
2324 ip
->i_d
.di_size
= pathlen
;
2327 * The inode was initially created in extent format.
2329 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
2330 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
2332 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
2333 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
2337 nmaps
= SYMLINK_MAPS
;
2339 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
2340 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
2341 &first_block
, resblks
, mval
, &nmaps
,
2348 resblks
-= fs_blocks
;
2349 ip
->i_d
.di_size
= pathlen
;
2350 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2352 cur_chunk
= target_path
;
2353 for (n
= 0; n
< nmaps
; n
++) {
2354 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
2355 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
2356 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
2357 BTOBB(byte_cnt
), 0);
2358 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
2359 if (pathlen
< byte_cnt
) {
2362 pathlen
-= byte_cnt
;
2364 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
2365 cur_chunk
+= byte_cnt
;
2367 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
2372 * Create the directory entry for the symlink.
2374 error
= xfs_dir_createname(tp
, dp
, link_name
, ip
->i_ino
,
2375 &first_block
, &free_list
, resblks
);
2378 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2379 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2382 * If this is a synchronous mount, make sure that the
2383 * symlink transaction goes to disk before returning to
2386 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2387 xfs_trans_set_sync(tp
);
2391 * xfs_trans_commit normally decrements the vnode ref count
2392 * when it unlocks the inode. Since we want to return the
2393 * vnode to the caller, we bump the vnode ref count now.
2397 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2401 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2402 xfs_qm_dqrele(udqp
);
2403 xfs_qm_dqrele(gdqp
);
2405 /* Fall through to std_return with error = 0 or errno from
2406 * xfs_trans_commit */
2408 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTSYMLINK
)) {
2409 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTSYMLINK
,
2412 DM_RIGHT_NULL
, link_name
->name
,
2413 target_path
, 0, error
, 0);
2423 xfs_bmap_cancel(&free_list
);
2424 cancel_flags
|= XFS_TRANS_ABORT
;
2426 xfs_trans_cancel(tp
, cancel_flags
);
2427 xfs_qm_dqrele(udqp
);
2428 xfs_qm_dqrele(gdqp
);
2430 if (unlock_dp_on_error
)
2431 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2442 xfs_mount_t
*mp
= ip
->i_mount
;
2446 if (!capable(CAP_SYS_ADMIN
))
2447 return XFS_ERROR(EPERM
);
2449 if (XFS_FORCED_SHUTDOWN(mp
))
2450 return XFS_ERROR(EIO
);
2452 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
2453 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
2455 xfs_trans_cancel(tp
, 0);
2458 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2459 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2461 ip
->i_d
.di_dmevmask
= evmask
;
2462 ip
->i_d
.di_dmstate
= state
;
2464 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2466 error
= xfs_trans_commit(tp
, 0);
2472 * xfs_alloc_file_space()
2473 * This routine allocates disk space for the given file.
2475 * If alloc_type == 0, this request is for an ALLOCSP type
2476 * request which will change the file size. In this case, no
2477 * DMAPI event will be generated by the call. A TRUNCATE event
2478 * will be generated later by xfs_setattr.
2480 * If alloc_type != 0, this request is for a RESVSP type
2481 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2482 * lower block boundary byte address is less than the file's
2491 xfs_alloc_file_space(
2498 xfs_mount_t
*mp
= ip
->i_mount
;
2500 xfs_filblks_t allocated_fsb
;
2501 xfs_filblks_t allocatesize_fsb
;
2502 xfs_extlen_t extsz
, temp
;
2503 xfs_fileoff_t startoffset_fsb
;
2504 xfs_fsblock_t firstfsb
;
2510 xfs_bmbt_irec_t imaps
[1], *imapp
;
2511 xfs_bmap_free_t free_list
;
2512 uint qblocks
, resblks
, resrtextents
;
2516 xfs_itrace_entry(ip
);
2518 if (XFS_FORCED_SHUTDOWN(mp
))
2519 return XFS_ERROR(EIO
);
2521 error
= xfs_qm_dqattach(ip
, 0);
2526 return XFS_ERROR(EINVAL
);
2528 rt
= XFS_IS_REALTIME_INODE(ip
);
2529 extsz
= xfs_get_extsz_hint(ip
);
2534 bmapi_flag
= XFS_BMAPI_WRITE
| (alloc_type
? XFS_BMAPI_PREALLOC
: 0);
2535 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
2536 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
2538 /* Generate a DMAPI event if needed. */
2539 if (alloc_type
!= 0 && offset
< ip
->i_size
&&
2540 (attr_flags
& XFS_ATTR_DMI
) == 0 &&
2541 DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
)) {
2542 xfs_off_t end_dmi_offset
;
2544 end_dmi_offset
= offset
+len
;
2545 if (end_dmi_offset
> ip
->i_size
)
2546 end_dmi_offset
= ip
->i_size
;
2547 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, ip
, offset
,
2548 end_dmi_offset
- offset
, 0, NULL
);
2554 * Allocate file space until done or until there is an error
2557 while (allocatesize_fsb
&& !error
) {
2561 * Determine space reservations for data/realtime.
2563 if (unlikely(extsz
)) {
2564 s
= startoffset_fsb
;
2567 e
= startoffset_fsb
+ allocatesize_fsb
;
2568 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
2570 if ((temp
= do_mod(e
, extsz
)))
2574 e
= allocatesize_fsb
;
2578 resrtextents
= qblocks
= (uint
)(e
- s
);
2579 resrtextents
/= mp
->m_sb
.sb_rextsize
;
2580 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2581 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
2584 resblks
= qblocks
= \
2585 XFS_DIOSTRAT_SPACE_RES(mp
, (uint
)(e
- s
));
2586 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
2590 * Allocate and setup the transaction.
2592 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2593 error
= xfs_trans_reserve(tp
, resblks
,
2594 XFS_WRITE_LOG_RES(mp
), resrtextents
,
2595 XFS_TRANS_PERM_LOG_RES
,
2596 XFS_WRITE_LOG_COUNT
);
2598 * Check for running out of space
2602 * Free the transaction structure.
2604 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2605 xfs_trans_cancel(tp
, 0);
2608 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2609 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
2614 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2615 xfs_trans_ihold(tp
, ip
);
2618 * Issue the xfs_bmapi() call to allocate the blocks
2620 xfs_bmap_init(&free_list
, &firstfsb
);
2621 error
= xfs_bmapi(tp
, ip
, startoffset_fsb
,
2622 allocatesize_fsb
, bmapi_flag
,
2623 &firstfsb
, 0, imapp
, &nimaps
,
2630 * Complete the transaction
2632 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2637 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2638 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2643 allocated_fsb
= imapp
->br_blockcount
;
2646 error
= XFS_ERROR(ENOSPC
);
2650 startoffset_fsb
+= allocated_fsb
;
2651 allocatesize_fsb
-= allocated_fsb
;
2654 if (error
== ENOSPC
&& (attr_flags
& XFS_ATTR_DMI
) == 0 &&
2655 DM_EVENT_ENABLED(ip
, DM_EVENT_NOSPACE
)) {
2656 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_NOSPACE
,
2659 NULL
, NULL
, 0, 0, 0); /* Delay flag intentionally unused */
2661 goto retry
; /* Maybe DMAPI app. has made space */
2662 /* else fall through with error from XFS_SEND_DATA */
2667 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2668 xfs_bmap_cancel(&free_list
);
2669 xfs_trans_unreserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
2671 error1
: /* Just cancel transaction */
2672 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2673 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2674 goto dmapi_enospc_check
;
2678 * Zero file bytes between startoff and endoff inclusive.
2679 * The iolock is held exclusive and no blocks are buffered.
2681 * This function is used by xfs_free_file_space() to zero
2682 * partial blocks when the range to free is not block aligned.
2683 * When unreserving space with boundaries that are not block
2684 * aligned we round up the start and round down the end
2685 * boundaries and then use this function to zero the parts of
2686 * the blocks that got dropped during the rounding.
2689 xfs_zero_remaining_bytes(
2694 xfs_bmbt_irec_t imap
;
2695 xfs_fileoff_t offset_fsb
;
2696 xfs_off_t lastoffset
;
2699 xfs_mount_t
*mp
= ip
->i_mount
;
2704 * Avoid doing I/O beyond eof - it's not necessary
2705 * since nothing can read beyond eof. The space will
2706 * be zeroed when the file is extended anyway.
2708 if (startoff
>= ip
->i_size
)
2711 if (endoff
> ip
->i_size
)
2712 endoff
= ip
->i_size
;
2714 bp
= xfs_buf_get_noaddr(mp
->m_sb
.sb_blocksize
,
2715 XFS_IS_REALTIME_INODE(ip
) ?
2716 mp
->m_rtdev_targp
: mp
->m_ddev_targp
);
2718 return XFS_ERROR(ENOMEM
);
2720 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
2721 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
2723 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, 1, 0,
2724 NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
2725 if (error
|| nimap
< 1)
2727 ASSERT(imap
.br_blockcount
>= 1);
2728 ASSERT(imap
.br_startoff
== offset_fsb
);
2729 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
2730 if (lastoffset
> endoff
)
2731 lastoffset
= endoff
;
2732 if (imap
.br_startblock
== HOLESTARTBLOCK
)
2734 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2735 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
2738 XFS_BUF_UNWRITE(bp
);
2740 XFS_BUF_SET_ADDR(bp
, xfs_fsb_to_db(ip
, imap
.br_startblock
));
2742 error
= xfs_iowait(bp
);
2744 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2745 mp
, bp
, XFS_BUF_ADDR(bp
));
2748 memset(XFS_BUF_PTR(bp
) +
2749 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
2750 0, lastoffset
- offset
+ 1);
2755 error
= xfs_iowait(bp
);
2757 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2758 mp
, bp
, XFS_BUF_ADDR(bp
));
2767 * xfs_free_file_space()
2768 * This routine frees disk space for the given file.
2770 * This routine is only called by xfs_change_file_space
2771 * for an UNRESVSP type call.
2779 xfs_free_file_space(
2787 xfs_off_t end_dmi_offset
;
2788 xfs_fileoff_t endoffset_fsb
;
2790 xfs_fsblock_t firstfsb
;
2791 xfs_bmap_free_t free_list
;
2792 xfs_bmbt_irec_t imap
;
2800 xfs_fileoff_t startoffset_fsb
;
2802 int need_iolock
= 1;
2806 xfs_itrace_entry(ip
);
2808 error
= xfs_qm_dqattach(ip
, 0);
2813 if (len
<= 0) /* if nothing being freed */
2815 rt
= XFS_IS_REALTIME_INODE(ip
);
2816 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
2817 end_dmi_offset
= offset
+ len
;
2818 endoffset_fsb
= XFS_B_TO_FSBT(mp
, end_dmi_offset
);
2820 if (offset
< ip
->i_size
&& (attr_flags
& XFS_ATTR_DMI
) == 0 &&
2821 DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
)) {
2822 if (end_dmi_offset
> ip
->i_size
)
2823 end_dmi_offset
= ip
->i_size
;
2824 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, ip
,
2825 offset
, end_dmi_offset
- offset
,
2826 AT_DELAY_FLAG(attr_flags
), NULL
);
2831 if (attr_flags
& XFS_ATTR_NOLOCK
)
2834 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
2835 /* wait for the completion of any pending DIOs */
2839 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_CACHE_SIZE
);
2840 ioffset
= offset
& ~(rounding
- 1);
2842 if (VN_CACHED(VFS_I(ip
)) != 0) {
2843 error
= xfs_flushinval_pages(ip
, ioffset
, -1, FI_REMAPF_LOCKED
);
2845 goto out_unlock_iolock
;
2849 * Need to zero the stuff we're not freeing, on disk.
2850 * If it's a realtime file & can't use unwritten extents then we
2851 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2852 * will take care of it for us.
2854 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
2856 error
= xfs_bmapi(NULL
, ip
, startoffset_fsb
,
2857 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
2859 goto out_unlock_iolock
;
2860 ASSERT(nimap
== 0 || nimap
== 1);
2861 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
2864 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2865 block
= imap
.br_startblock
;
2866 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
2868 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
2871 error
= xfs_bmapi(NULL
, ip
, endoffset_fsb
- 1,
2872 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
2874 goto out_unlock_iolock
;
2875 ASSERT(nimap
== 0 || nimap
== 1);
2876 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
2877 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2879 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
2880 endoffset_fsb
-= mod
;
2883 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
2885 * One contiguous piece to clear
2887 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
2890 * Some full blocks, possibly two pieces to clear
2892 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
2893 error
= xfs_zero_remaining_bytes(ip
, offset
,
2894 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
2896 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
2897 error
= xfs_zero_remaining_bytes(ip
,
2898 XFS_FSB_TO_B(mp
, endoffset_fsb
),
2903 * free file space until done or until there is an error
2905 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2906 while (!error
&& !done
) {
2909 * allocate and setup the transaction. Allow this
2910 * transaction to dip into the reserve blocks to ensure
2911 * the freeing of the space succeeds at ENOSPC.
2913 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2914 tp
->t_flags
|= XFS_TRANS_RESERVE
;
2915 error
= xfs_trans_reserve(tp
,
2917 XFS_WRITE_LOG_RES(mp
),
2919 XFS_TRANS_PERM_LOG_RES
,
2920 XFS_WRITE_LOG_COUNT
);
2923 * check for running out of space
2927 * Free the transaction structure.
2929 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2930 xfs_trans_cancel(tp
, 0);
2933 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2934 error
= xfs_trans_reserve_quota(tp
, mp
,
2935 ip
->i_udquot
, ip
->i_gdquot
,
2936 resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
2940 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2941 xfs_trans_ihold(tp
, ip
);
2944 * issue the bunmapi() call to free the blocks
2946 xfs_bmap_init(&free_list
, &firstfsb
);
2947 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
2948 endoffset_fsb
- startoffset_fsb
,
2949 0, 2, &firstfsb
, &free_list
, NULL
, &done
);
2955 * complete the transaction
2957 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2962 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2963 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2968 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
2972 xfs_bmap_cancel(&free_list
);
2974 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2975 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
2981 * xfs_change_file_space()
2982 * This routine allocates or frees disk space for the given file.
2983 * The user specified parameters are checked for alignment and size
2992 xfs_change_file_space(
2999 xfs_mount_t
*mp
= ip
->i_mount
;
3004 xfs_off_t startoffset
;
3009 xfs_itrace_entry(ip
);
3011 if (!S_ISREG(ip
->i_d
.di_mode
))
3012 return XFS_ERROR(EINVAL
);
3014 switch (bf
->l_whence
) {
3015 case 0: /*SEEK_SET*/
3017 case 1: /*SEEK_CUR*/
3018 bf
->l_start
+= offset
;
3020 case 2: /*SEEK_END*/
3021 bf
->l_start
+= ip
->i_size
;
3024 return XFS_ERROR(EINVAL
);
3027 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
3029 if ( (bf
->l_start
< 0)
3030 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
3031 || (bf
->l_start
+ llen
< 0)
3032 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
3033 return XFS_ERROR(EINVAL
);
3037 startoffset
= bf
->l_start
;
3041 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3043 * These calls do NOT zero the data space allocated to the file,
3044 * nor do they change the file size.
3046 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3048 * These calls cause the new file data to be zeroed and the file
3049 * size to be changed.
3051 setprealloc
= clrprealloc
= 0;
3054 case XFS_IOC_RESVSP
:
3055 case XFS_IOC_RESVSP64
:
3056 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
3063 case XFS_IOC_UNRESVSP
:
3064 case XFS_IOC_UNRESVSP64
:
3065 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
3070 case XFS_IOC_ALLOCSP
:
3071 case XFS_IOC_ALLOCSP64
:
3072 case XFS_IOC_FREESP
:
3073 case XFS_IOC_FREESP64
:
3074 if (startoffset
> fsize
) {
3075 error
= xfs_alloc_file_space(ip
, fsize
,
3076 startoffset
- fsize
, 0, attr_flags
);
3081 iattr
.ia_valid
= ATTR_SIZE
;
3082 iattr
.ia_size
= startoffset
;
3084 error
= xfs_setattr(ip
, &iattr
, attr_flags
);
3094 return XFS_ERROR(EINVAL
);
3098 * update the inode timestamp, mode, and prealloc flag bits
3100 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
3102 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
3105 xfs_trans_cancel(tp
, 0);
3109 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3111 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3112 xfs_trans_ihold(tp
, ip
);
3114 if ((attr_flags
& XFS_ATTR_DMI
) == 0) {
3115 ip
->i_d
.di_mode
&= ~S_ISUID
;
3118 * Note that we don't have to worry about mandatory
3119 * file locking being disabled here because we only
3120 * clear the S_ISGID bit if the Group execute bit is
3121 * on, but if it was on then mandatory locking wouldn't
3122 * have been enabled.
3124 if (ip
->i_d
.di_mode
& S_IXGRP
)
3125 ip
->i_d
.di_mode
&= ~S_ISGID
;
3127 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3130 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
3131 else if (clrprealloc
)
3132 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
3134 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3135 xfs_trans_set_sync(tp
);
3137 error
= xfs_trans_commit(tp
, 0);
3139 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);